Improving NPS using Multi-Modal Recommendation Engine

Problem

Clinicians Team at LYBL were facing the problem of too much cognitive load while doing multiple back to back consultations in a day.

They had to:

  • Decode user symptoms

  • Infer possible root causes

  • Pick the right intervention strategy

  • Then explain all this - in 30 minutes or less

The experience for users?

  • Sometimes felt surface-level

  • Took too long to get to the real issue

  • Missed the opportunity to personalize from Day 1


The Dual-Engine Solution

Marrying Pathology Intelligence with Generative AI

Helping health experts deliver faster, deeper, and smarter consultations

We built a hybrid recommendation engine combining:

Pathology Intelligence

  • Every user’s inputs were mapped to a Pathology ID:
    a structured model of likely root causes (e.g. sluggish liver, gut imbalance, hormonal dysregulation)

  • These mappings weren’t generic - they were built with clinicians, based on thousands of cases

  • The engine deprioritized symptom chasing, and focused on core disruptions in body systems

AI Layer for Personalization & Presentation

  • On top of this clinical intelligence, we built an AI engine that:

    • Pulled in user context (mood, behavior logs, content viewed)

    • Generated a narrative summary for the clinician

    • Recommended talking points, nudges, and wellness content

What I Led

I led the creation of a multi-modal engine that used pathology mapping and AI to help clinicians uncover root causes — faster, deeper, and more personalized than ever before.

  • Defined the Pathology ID mapping framework with health experts

  • Designed the dual-layer recommendation flow (Pathology logic + AI prompts)

  • Created a tagging system for personalized recommendations

  • Oversaw integration into the clinician dashboard - with real time sync

  • Prioritized trust + transparency - AI was always explainable and editable

Impact

Area

Result

User NPS

+18% after engine rollout

Expert prep time

-35% average reduction

Plan relevance (expert-rated)

From 3.2 → 4.6 stars

Diagnosis Accuracy

86% match accuracy in early mapping

This wasn’t just AI, and it wasn’t just clinical mapping.
It was tech that respected the body’s complexity - and experts’ time.

We didn’t automate decision-making. We automated clarity.